OpenCV

OpenCV now has official merchandise for the first time! Quality t-shirts, sweaters, pet bandanas, and more are available in the official shop. Get an "I Love OpenCV" shirt in English, Spanish, and simplified Chinese. Shop now: https://lnkd.in/dSPNRCjE #OpenCV #ComputerVision #AI #OpenSource #NonProfit #ILoveOpenCV

SAMURAI - an advance in video segmentation and tracking! ?? ?? ?? I tested SAMURAI on a complex and crowded scene, and the results look promising. Would you like to see a YouTube stream covering this model? SAMURAI is research released by Hsiang-Wei H., Cheng-Yen Yang, Wenhao Chai, Cheng-Yen Yang, Zhongyu Jiang, and Jenq-Neng Hwang from the University of Washington, aiming to enhance the visual object tracking capabilities of the Segment Anything Model (SAM2). It addresses SAM2's limitations in handling crowded scenes and managing occlusions. - Motion Modeling: The researchers employ a Kalman Filter to predict the object's position and dimensions, which helps select the most confident mask from multiple candidates. - Motion-Aware Memory Selection: Instead of the original fixed-window memory approach, the researchers introduced a selective method that considers the mask affinity score, object occurrence score, and motion score to choose the relevant frames from the memory bank. Links to the code, paper, and my dirty Google Colab are in the comments below ???? #computervision #opensource #segmentation #tracking

It is impossible to imagine any modern airport without computer vision systems — AI, among other things, helps sort luggage with almost no errors, even millions of pieces of luggage per month. https://lnkd.in/d5KS--m3

Thank you to everyone who came to play with us at EdTech Taiwan! We had a lot of fun, maybe too much fun at the show. #EdTech #XR #VR #AR

??Using DepthPro and YOLO for Depth Measurement and 3D Reconstruction?? Distance + Depth + Point Cloud + 3D Bounding Box I’ve been working on integrating the DepthPro model with YOLO segmentation to create a system for metric distance measurement, depth estimation, and 3D point cloud reconstruction with object segmentation. The goal is to develop an efficient and accurate setup for capturing depth and spatial relationships in a single process. DepthPro is a powerful tool for monocular depth estimation, producing high-resolution depth maps that allow us to measure distances without needing detailed camera data. Combined with YOLO for object detection, this approach enables the generation of 3D point clouds with bounding boxes around objects — bringing us closer to creating monocular systems for use in autonomous vehicles. Big thanks to Nicolai Nielsen and TAI DO for their inspiring ideas on reconstruction and distance estimation that helped shape this approach. Looking forward to seeing where this technology can take us in the field of 3D vision! #ComputerVision #3DReconstruction #DepthEstimation #AI #MachineLearning

The annual conference for the Robot Operating System (ROS) recently took place in Odense, Denmark, and on OpenCV Live we've got Katherine Scott (DevRel for ROS and more) to tell us all about it. Tune in to see the latest in robotics simulation, visualization, AI and hardware. #OpenCV #ComputerVision #AI #ROS #ROSCon #OpenSource #Hardware #Robotics

The software we use everyday doesn't leverage the power of sight. But it will soon. Roboflow has raised an additional $40M to further grow open source and enterprise visual AI. Delivering on the promise of computer vision and visual AI is bigger than any one co. Open source packages we maintain with the community see millions of downloads. Roboflow Universe supports 500M+ user created and shared images across 150k+ models. The hosted platform has delivered 1M hours of GPU training to open source projects. We remain committed to being the starting point for those tinkering, creating, and scaling up your computer vision projects from dataset to model to deployment (and, increasingly, in reverse order). We've also seen rapid adoption from 25,000 organizations, including some of the world's largest Fortune 100 companies, operating at worldwide scale. Roboflow's solutions keep trains on the tracks, ensure your packages show up on time, and provide the building materials for your homes. 40% of global GDP (!) is wrapped up in these operationally intensive, real world businesses, and they're seeing millions in ROI partnering with Roboflow. We'd love to build with you.

Rerun 0.20 comes with H.264 support ???? The second part of video support in Rerun is here: support for H.264 video files in the native viewer. Video is hard, so we ended up building a bunch of debugging tools for ourselves along the way. Some of those are now exposed in the selection panel in the Rerun viewer to help you understand your video data better. The next piece after this will be support for sending individual video encoded frames at a time to Rerun without the need for a container format like mp4. This will then pave the way to live ingest video streams as they are encoded. Check out the full release blog post (?? link in the comments)

Ultra-light scene recognition — sometimes we just need a little more insight into the environment and don't have a lot of compute to spare. No problem — we can customize scene recognition to run in ms on your edge device! https://lnkd.in/dU7SVMFC

Fresh off our blog post, let's take a look at what advanced self-driving car technologies are working now (and how these technologies will work in the near future). https://lnkd.in/dJ4wNpp7